The cost of the next unit is often the only number that matters.
The Idea
Average cost is total cost divided by the number of units. Marginal cost is what one more unit adds. When most of a cost is fixed, such as a platform fee or a data center, the two can be far apart: the average stays high while the next unit costs almost nothing. When a fixed cost is about to step up, the gap runs the other way. Decisions about growth, reclamation, and pricing should usually be made at the margin.
Economists have argued about which number to price on for nearly a century. In 1938, Harold Hotelling proposed that railways and utilities charge only marginal cost, with taxes covering their large fixed costs. Ronald Coase’s 1946 reply identified the catch: if nobody pays the fixed cost directly, nobody learns whether users value the service enough to cover it. He favored multipart pricing, combining a charge for each unit used with a share of the fixed cost. Platform fees plus per-unit charges, in vendor contracts and internal chargeback models alike, follow the same logic.
In ITAM Terms
Your average cost per user for a platform might be $40 a month once you spread the platform fee, support, and admin across everyone. But user 1,001 might cost $12, or nothing if you’re under a tier ceiling, or $50,000 if it pushes you into the next band. Knowing where you sit on that curve changes the conversation.
Most unit metrics are averages. Cost per user or per transaction over a quarter shows a trend, and the FinOps Foundation’s introduction to cloud unit economics shows how teams use such metrics to tie spend to business value. The marginal view answers a different question: what will the next decision cost, or save?
Where You’ll See It
- Software: Tiered and banded pricing creates cliffs. Reclaiming 30 seats might drop you below a threshold and save far more than 30 times the per-seat price.
- Hardware: One more VM on an underused host costs little. The VM that forces a new host, more per-core licenses, and another support contract costs a lot.
- Cloud: Unit economics (cost per transaction, per customer) is how FinOps teams link spend to business activity, and tiered rates change what each additional unit costs. In billing data that follows the FOCUS specification, the SKU Price ID column supports analysis by pricing tier and tracks unit price changes over time.
- AI: Seat licenses hid marginal cost; per-token pricing puts it on every request. The Tokenomics Foundation’s Big-T Notation shows why the next request has no fixed price: hidden model calls and nested agents can multiply the tokens behind it. The FinOps Foundation’s paper on token costs in SaaS adds that the infrastructure around a model can account for 40 to 60 percent of an AI feature’s spend, so the token price alone understates what each feature costs to run.
Governance Takeaway
Report both numbers. Average cost supports chargeback and benchmarking; marginal cost supports the actual decision to add, remove, or reclaim. When you design chargeback, consider Coase’s approach: a fixed share of the platform plus a usage charge that reflects marginal cost. Consumers then see both what the platform costs to keep and what their next unit costs to use.
Try This
For one major contract, map the pricing bands and mark where your current count sits. How close are you to the next cliff, in either direction?
